Nvidia used Jensen Huang’s two-day trip to Tokyo this week to introduce Cosmos 3 Edge. The new model is a compact version of Nvidia’s physical AI system, built to handle navigation and perception for robots and vision systems in real time. It extends the open Cosmos 3 foundation model Nvidia introduced in May. The release lands alongside a coalition of Japanese industrial firms that Nvidia says plans to include Kawasaki Heavy Industries, Hitachi, and Fujitsu. For Nvidia, physical AI is becoming the answer to what comes after the data center chatbot boom: models small enough to run inside a robot, a hospital device, or a factory line.
Huang framed the moment in sweeping terms in a Wednesday statement. “The next frontier of AI is in the physical world, and this is a once-in-a-generation opportunity for Japan,” he said. Japan “invented modern manufacturing,” he added, and now has the opportunity “to reinvent it for the age of intelligent industries.”
CNBC reported that Japan intends to purchase 27,500 of Nvidia’s next-generation Rubin chips to build a homegrown foundation model for robots. That figure gives the announcement a hardware anchor the coalition itself lacks. Kawasaki Heavy Industries, Hitachi, and Fujitsu carry only that “intend to join” language attached to their names so far. None of the three has signed deployment terms.
The push follows a Microsoft commitment made earlier this year. Microsoft put $10 billion into Japanese AI infrastructure and cybersecurity. It also runs alongside SoftBank’s continued buildout, with partners including Microsoft and Sakura Internet.
The U.S. International Trade Administration puts Japan’s AI market at $27.9 billion by 2029. That number explains why Nvidia, Microsoft, and SoftBank are all placing bets in the same market at once. Barclays research chief Ajay Rajadhyaksha said last month that Japan’s diverse AI and clean-energy growth stories give it a structural edge over other Asian economies, a comment he made to CNBC before this week’s coalition news broke.
Nvidia is also expanding into Japan’s healthcare sector through Tokyo-1, an AI drug discovery consortium run by Xeureka, a subsidiary of trading house Mitsui. The consortium has been active since 2023. It runs on Nvidia’s BioNeMo Agent Toolkit, software built to automate drug discovery workflows. Its users now include Ono Pharmaceutical, Astellas Pharma, and Daiichi Sankyo. Kawasaki Heavy Industries is working with Nvidia separately on industrial automation, a narrower deal than its role in the broader coalition.
The sequence itself is worth noting. Cosmos 3 launched openly in May as a downloadable foundation model. Cosmos 3 Edge arrives two months later bundled to named regional partners instead. Nvidia’s announcement does not include a deployment timeline for the edge model. It also omits usage figures from Fujitsu, Hitachi, or Kawasaki Heavy Industries beyond their stated intent to join.
Robotics and vision AI teams should benchmark Cosmos 3 Edge against existing edge-inference stacks before committing budget to it as a default choice. For anyone tracking Nvidia’s push beyond data center GPUs, the 27,500-chip Rubin order is the more verifiable number until the coalition produces adoption figures of its own.
According to CNBC’s July 16, 2026 report by Jenny Lee, Nvidia unveiled Cosmos 3 Edge and its Japan coalition during Jensen Huang’s Tokyo visit.
